@Marvin I need the prefect worker spawned through

prefect worker start -t docker --pool <pool-name>

to use the underlying gpu... How can I accomplish that? Is this something I would need to add to the docker image or to the prefect worker command?

To enable GPU support for Prefect workers spawned through

prefect worker start

with the Docker task runner, you would need to make certain configurations to your Docker environment rather than the Prefect worker command directly. Here are the steps you'd typically need to follow: 1. Ensure GPU Support in Docker: You'd have to make sure that the Docker daemon and your Docker image can access the host's GPU resources. This often involves installing NVIDIA drivers and NVIDIA Container Toolkit on your host machine. 2. Configure Docker Image: Your Docker image needs to be configured to support GPU usage. This could mean including necessary libraries and making sure it is built upon an appropriate base image, such as one from NVIDIA's NGC which comes pre-configured with CUDA and GPU support. 3. Modify Prefect Worker Configuration: When specifying your worker configuration, you may need to pass additional arguments to Docker to enable GPU access, such as

--gpus all

, although such options are not passed via the

prefect worker start

command directly. Instead, this would likely involve customizing your Docker execution environment or entrypoint in the Dockerfile. The documentation excerpt retrieved doesn't explicitly mention running Prefect with GPU support, but for serverless GPU jobs, it recommends Google Vertex AI, indicating that custom configuration or a tailored environment setup is required for specific types of compute resources. For a comprehensive guide or specific code examples, you might need to consult more detailed resources or community advice on how to configure Docker for GPU with Prefect workers. Prefect Serverless Workers Guide If you need further assistance, I'd be happy to help you delve deeper with a sigh.